Skip to content

Posterior-Mean Rectified Flow

An image restoration method minimizing mean squared error

Visit official site

What Posterior-Mean Rectified Flow is

PMRF is described as a restoration algorithm that pairs posterior mean prediction with a rectified flow model to minimize mean squared error.

What you can do with it

  • Reduce noise in a degraded image
  • Add color to a monochrome image
  • Raise resolution through super resolution

How restoration works

  • A posterior mean is predicted in the first stage
  • Two training stages minimize a simple MSE loss
  • A rectified flow model turns the prediction into an image

Pros & cons

✓ What we like

  • Handles several kinds of image degradation
  • A rectified flow model produces high quality output

! What to watch out for

  • Code and pretrained models must be obtained
  • A degraded image has to be prepared first

FAQ

What is optimized first?

The posterior mean prediction is optimized to minimize mean squared error.

Which tasks are listed?

Denoising, colorization, inpainting and super resolution.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review